verifiedhigh · 0.90updated 2026-10-04

PagedAttention: Efficient Memory Management for LLM Serving

paper

Fields

sources/src5-paged-attention-paper

URL

https://arxiv.org/abs/2309.06180

Publisher

UC Berkeley

Kind

paper

Published

2023-09-01

Accessed

2026-10-04

notes

PagedAttention enables efficient KV cache management by partitioning into fixed-size blocks. Eliminates memory fragmentation. Key innovation behind vLLM. Foundation for modern LLM serving systems.

Notes

PagedAttention enables efficient KV cache management by partitioning into fixed-size blocks. Eliminates memory fragmentation. Key innovation behind vLLM. Foundation for modern LLM serving systems.

Cross-references

This record references no other records.